The nature of most problems addressed by robotics requires that roboti
c systems possess real-time properties. Additionally, as a result of s
teady increases in power and decreases in the cost of technology, it h
as become feasible to integrate sophisticated vision systems into robo
tic tasks. This can be seen by the recent interest in active vision. T
he purpose of this paper is two-fold: we first present a novel archite
cture for real-time active vision systems, and then enhance the archit
ecture with a unified approach to fault tolerance. Our system is desig
ned modularly in order to enable the flexible addition of hardware and
software redundancy and also to allow reconfiguration when and where
needed. This gives us the ability to handle faults in the context of a
ctive vision. Further, the distribution of software on the available h
ardware is such that users can utilize a dual-mode of execution (simul
ation and rapid prototyping). Lastly, the tests we ran on the implemen
ted architecture in order to validate the results of the experimentati
on and simulations show a good correlation of parameters. (C) 1998 Aca
demic Press Limited.